a guide to season ticket holder management · a guide to season ticket holder management version...

AMP Series – Academia Meets Practice [Mike Lewis & Manish Tripathi]

Note on Season Ticket Management

[email protected] CONFIDENTIAL [email protected]

Page | 1

A Guide to Season Ticket Holder Management

Version 1.0: October 2014

Over the past year, we have been engaged in a variety of efforts centered in the sports

marketing space. We publish a blog that comments on newsworthy topics (e.g. The Ed

O’Bannon case or the Washington Redskins controversy) and we have developed two

sports analytics oriented courses that are taught at Emory University.

In the next stage of our evolution, we are launching a series that we are calling Academia

Meets Practice (AMP). The goal of the series is to present concepts, ideas and techniques

that may be useful to key decision makers at sports organizations and students aspiring to

careers in sports marketing. In general, the focus will be on methods for using data to

improve marketing and business decisions. Over the last decade, there has been

explosive growth in the field of marketing analytics. Given the nature and amount of

customer data they possess, sports organizations are in many ways well positioned to

leverage these techniques.

The purpose of this particular article is to present a structure for Customer Relationship

Management (CRM) in the sports industry. Fan management has much in common with

customer management in traditional marketing contexts, but there are a few unique

aspects that create special challenges for sports marketers. In this document we review

these challenges and discuss potential solutions. Our goals for this article include:

GOALS

Establish the value of using Customer Lifetime Value as the underlying objective

of season ticket buyer management.

Introduce the basic concepts needed for developing a statistically-based decision-

support system focused on season ticket holder management. Issues covered

include modeling customer acquisition, customer retention, quantity decisions and

migration between ticket quality levels.

Highlight the importance of considering customer expectations in season ticket

holder management. We also provide some preliminary insights into how the

analyst can include “expectations” in statistical models of customer behavior.

http://scholarblogs.emory.edu/esma




Page | 2

Discuss the complexities in dealing with ticket prices in a decision-support system.

This discussion highlights various sources of information that may be used to

develop a ticket quality index.

Development of a framework that may be used to create / calculate dynamically

optimal season ticket management policies.

In terms of organizational structure, we begin with a discussion of the appropriate CRM

objective for sports franchises. The specification of an objective is an important, but

frequently skipped, step in the design of CRM systems.

We then discuss retention modeling. While season ticket holder retention is just one of

many important fan decisions, the retention decision is likely at the heart of any

relationship marketing decision-support system.

Following the discussion of retention, we focus on how teams provide value to fans and

how the structure of this “value” may lead to complex, dynamic decision-making on the

part of fans. We also briefly discuss modeling techniques for considering decisions

related to ticket quality, ticket quantity and package size.

Finally, the paper also includes material related to the conversion of customer response

model results to marketing policies. This is an important but challenging topic as it

requires a holistic view of the organization, marketing insights and sophisticated analysis

tools.

Sections

1. Discussion of Fan Management Objectives

2. Fan Behavior: Customer Retention and Acquisition

3. Fan Behavior: Extensions

4. Ticket Quality and Pricing

5. Fan Expectations

6. Optimal Fan Management

7. Final Comments




Page | 3

1. Fan Management Objectives

Sports franchises operate in an environment where it is increasingly possible to track

multiple elements of an individual customer’s behavior. For example, season ticket

contracts allow tracking of year over year retention, and bar codes allow tracking of

ticket and concession usage. This type of detailed customer retention and revenue data

makes it possible for sports franchises to use customer-focused metrics such as Customer

Lifetime Value (CLV) and Customer Equity (CE) as primary marketing goals.

CLV is a core CRM metric that attempts to put a dollar value on a firm’s customer

relationship assets. The idea behind CLV is that individual customers can be viewed in

terms of the value or profit that they will contribute to the firm over some time period.

For example, a customer that purchases a full MLB season ticket package (81 games) for

two $100 tickets would provide a baseball team with $16,200 in revenue per season. If

that customer was retained for 10 years, the customer’s lifetime revenue value would be

$162,000 (assuming no inflation or discounting). If the customer was only retained for 5

years this lifetime revenue drops to $81,000.

While the previous example is obvious, there is a critical point to be made in regards to

retention rates. Retention rates are much like compound interest as the impact of small

changes becomes large over time. If a team has a retention rate of 95% per year the

probability that a customer will last 10 years is 60%. In contrast, if the retention rate is

90% the probability of a customer being retained for 10 years drops to just 35%. Given

that teams have high fixed but low marginal costs of serving a customer this means that

small changes in retention rates can greatly impact profits.

It is instructive to express CLV in equation form. A simple formula for CLV is given in

equation (1).

(1) ∑

In this formula Pr(Retentiont) is the probability that the customer is retained in period t,

Revenuet is the revenue produced by the customer in period t, Costt is the cost to serve

the customer in period t, and T is the number of periods used for the CLV calculation.

The basic CLV calculation can be extended to include additional factors such as inflation

or discounting. The equation highlights the role of retention in the value of customer

assets. In words, the equation just says that lifetime value is the sum of expected




Page | 4

profits generated from a specific customer over time. A critical insight is that the

probability of retention is an incredibly important element of customer value.

However, the structure of the CLV equation as written in equation (1) should give sports

marketing executives pause. The issue is that the equation is written without regard to

marketing or team decisions. If CLV is not a function of marketing the implication is

that relationship marketing doesn’t drive customer decisions. Below we rewrite the

equation to make explicit that retention, revenue and costs may be a function of team

marketing decisions at time t, Mt.

(2) ∑

For the purposes of this document, we take a very broad view of what constitutes

“marketing.” In particular, a controversial issue is how to treat team quality. Our view is

that team decisions related to payroll and roster construction should ultimately be viewed

in terms of marketing consequences. This statement may be provocative as purists may

believe that winning should be the ultimate goal and marketing and revenue generation

are more appropriately viewed as supporting activities.

However, there is a strong case to be made for considering team payroll and roster

decisions as marketing activities. From a consumer decision-making perspective,

winning and losing are likely the main drivers of fan interest (in the near term). Winning

rates are an interesting variable from a business perspective, in that sports teams actually

have an objective and observable measure of quality. Given the established correlations

between both payroll and winning, and winning and attendance, investments in team

quality through payroll represent a direct means for driving attendance.

Investment in payroll and winning also has a longer term benefit. In our studies of brand

equity in professional and college sports, we have consistently found that the key to

creating a valuable, loyalty- inducing brand is winning championships. The implication

of this finding is that investments in payroll and the pursuit of championships are often

investments in brand equity and fan loyalty.

Writing the equation for Customer Lifetime Value as a function of marketing is useful as

it makes explicit that the role of marketing (and payroll decisions) is to increase the value

of customer assets. A frequent mistake in the field of CRM is to simply treat CLV as a

fixed quantity that may be used to segment customers into groupings that vary based on

expected CLV. This is problematic because it implicitly assumes that CLV is

independent of marketing decisions. In the sports context, this is a particularly bad

http://scholarblogs.emory.edu/esma




Page | 5

assumption as investments in team quality (payroll) are obviously linked to season ticket

holder retention and willingness to pay.

One final point before we move on to the key principles for this section. We have not

placed much emphasis on data integration issues. This may be a significant issue in

CRM. For example, in the preceding discussion we did not consider the possible

complexities in measuring customer revenues. Revenues can include tickets, parking,

merchandise, concessions, and potentially many other factors. As these revenues are

collected through different channels it may be difficult to obtain a clear picture of

individual level revenues. We do not consider these data integration issues in this note.

Key Principles

The first step in customer management should be to set a clear objective. The

objective we advocate is to maximize the lifetime value of current and prospective

customers.

Customer Lifetime Value (CLV) should be viewed as a function of marketing

decisions. When this is the case, the marketing function becomes more

accountable and goal-oriented. The key point is that the marketing department’s

goals are related to the creation and management of valuable assets (customer

relationships) rather than to short-term objectives.

Marketing assets, such as customer relationships and brand equity, are a function

of team payroll decisions and team performance. Team performance should not

be viewed as unrelated to marketing policy.




Page | 6

2. Retention and Acquisition

In the last section, we made the point that customer retention is a key driver of customer

lifetime value (CLV). The obvious implication is that a primary objective for teams

should be to understand what drives retention decisions. When this is accomplished then

teams can better invest in marketing programs or customer specific interventions.

Understanding the drivers of retention is best accomplished through the development of

predictive models of season ticket holder retention. Perhaps the simplest decision to be

modeled is whether or not an individual customer will choose to renew his / her season

tickets. We will forgo consideration of ticket quality or ticket quantity for now.

We can begin with a simple equation along the lines of the following:

In words, this equation says that the probability (Pr) of renewal is a function (f) of

something. The something is where things get interesting. In general, we might think

that the “something” will include things about the team, the customer and the marketing

decisions of the organization. From a conceptual perspective, the “something” should be

focused on the decisions that the club can make. This is important because the end goal

of this type of work needs to be some type of decision-support system that allows the

club to understand how its decisions impact its customers’ decisions. From a practical

perspective, CRM systems and predictive analytics are often constrained by data

availability. In other words, we often predict based on the data we have rather than the

data we truly want.

Dependent Variable. For the analysis of retention, the dependent variable (the “what”

we want to predict) is the Yes / No renewal decision made by season ticket holders. This

decision may be treated as a binary outcome. Another Yes / No decision that is common

in CRM applications is customer acquisition. We briefly comment on modeling this

decision later in this section.

Explanatory Variables. The second type of variable that we need to consider is

explanatory variables. As the name suggests these are the variables that we will use to

explain the yes / no decision. These explanatory variables can come from a variety of

sources and may be under the control of different decision makers. For instance, one

team-level factor that may impact customer retention is team success. Fans clearly are




Page | 7

more interested in supporting winning teams than losing teams. Winning rates are an

interesting variable to include in a marketing decision support system since team quality

is probably beyond the purview of the marketing department. Other potential team level

factors to include in a retention model are given below.

Team-Level Variables:

Last season’s winning percentage

Last season’s post season results

Previous season finish (place or games back)

Team payroll (to control for star power)

Number of all-stars

Any other team factors that affect customer interest or expectations for next season

At the core of most CRM systems is individual level customer data. Across many

industries a consistent finding is that past loyalty is the best predictor of future loyalty.

Many categories use recency (time since last purchase), frequency (the number of

purchases) and monetary value (the cumulative amount spent) as core predictors of future

customer loyalty. These RFM measures are derived from customer transaction history

data. In addition to basic transaction history measures, teams may find it useful to model

retention as a function of other customer data such as distance to stadium or amount spent

on concessions. The relevant customer data is something of an empirical question.

Teams should start broadly and use statistical techniques to determine what customer

traits are useful for predicting future purchases.

Customer-Level Factors:

Time as a season ticket holder

Attendance frequency (if available)

Distance to the stadium / arena

Transaction history measures (ticket prices paid, etc.)

Customer initiated interactions with customer service

Another type of variable that needs to be considered is “marketing decisions.” These are

the variables over which that the marketing department has direct control. These are also

the critical variables for understanding the pay-off to specific marketing programs. These




Page | 8

marketing decision variables include the traditional elements of the marketing mix such

as prices, discounts, advertising and customer contacts.

Marketing Decisions:

Price

Customer contacts (frequency and type)

Promotions

Discounts

The preceding list is in no way meant to be exhaustive. The appropriate variables to

collect and study are something of an empirical question. Teams should weigh the cost

of data collection and experiment with variables to ascertain which are the most

predictive.

Modeling Customer Retention.

The move from data collection to analysis is often a significant challenge. Many

organizations still exist in the realm of descriptive statistics or simple cross-tab type

relationships. For example, a team might do a simple comparison of retention rates as a

function of average prices in each section. These types of relationships are easily

visualized and can be of significant value. However, as the preceding discussion about

data suggests, there may be a significant number of factors that affect decisions. It is for

this reason that we advocate for the use of statistical models (that can separate out

multiple effects) to understand the relationships between factors.

The fact that our decision is a yes or no choice rather than a quantity may complicate the

modeling process. The problem is that the standard linear regression model is designed

to predict continuous variables (like income) rather than dichotomous outcomes (such as

yes or no renewal decisions). Specifically, the main objection to a linear regression

model is that the model predictions are not constrained to be between zero and one

(remember we are trying to predict a probability).

For example, we might wish to predict the probability of a purchase (or renewal) based

on price, years as a customer and the team’s winning percentage last season. Equation

(3) listed below illustrates how these factors might look in a linear regression model.

(3)




Page | 9

For purposes of illustration, let’s say that the above equation was estimated using data on

season ticket holder behavior and the resulting coefficients are given in equation (4).

(4)

Equation (4) may be viewed as both an analysis of customer behavior and also as a

decision support tool. For example, the coefficients suggest that a $10 increase in price

reduces retention by 10% while an increase of winning percentage of 10% increases

retention by 12%. This type of information can begin to provide insight into how

management decisions impact fan response.

An alternative to the linear probability model is logistic regression. Logistic regression

shares some similarities with linear regression, but there are important differences. In

terms of similarities, the logistic regression can also predict the likelihood of an event as

a function of a set of explanatory variables. Just as in the preceding example, retention

could be predicted as a function of price, years of purchase and winning percentage.

The primary benefit of logistic regression is that this model constrains the estimated

probabilities to between 0 and 1. This makes the model’s predictions more interpretable

and results in more intuitive “scores” for each customer in the database. However, the

logistic regression model does involve some additional complexity. First, the logistic

regression requires specialized software such a SAS, R or SPSS. Second, interpretation

of coefficients is not as straight-forward. Equation (5) presents the expression required to

predict probabilities based on estimates from a logistic regression.

(5)

( )

Customer Acquisition. Thus far we have focused on retention or repeat buying. This is

a yes or no decision: do I renew my tickets or do I allow my season tickets to lapse?

Another yes/no decision made by customers is the initial decision to become a season

ticket holder. Linear probability models or binary logistic models based on individual,

team and marketing variables may also be used to analyze the drivers of customer

acquisition.

There are a couple of items that separate retention and acquisition analysis. The primary

issue is that in the case of acquisition the firm is unlikely to have a great deal of customer

specific data. A particularly important type of acquisition study is the analysis of

acquisition campaigns. For example, a team might use direct mail to offer an

introductory customer discount to a set of prospects. Logistic regression could be used to




Page | 10

determine how multiple factors such as offered discount, distance from stadium or arena

or demographics based on zip code impact acceptance of an offer. This type of

information could be used to refine future direct marketing efforts.

Key Principles

The development of a statistical model of retention has multiple benefits. The

modeling exercise reveals what factors have a significant impact on consumer

decision making and what factors do not drive customer behavior. For example,

the model might reveal that concession prices impact retention while parking

prices do not.

In addition, the relationship between team actions and fan decisions becomes

quantifiable. For example, a model could reveal that when winning percentage

increase by 1% that retention increases by 2%.

Retention (acquisition) analysis may be accomplished using relatively simple tools

such as linear regression performed in an excel spreadsheet or with slightly more

complicated procedures such as logistic regression.




Page | 11

3. Analyzing Fan Behavior: Extensions

As noted above, season ticket-holder behavior is more complicated than simple yes or no

renewal decisions. Season ticket holders also decide on ticket quality, size of package

and number of tickets per game. Data is also increasingly available about attendance

decisions and reselling behavior at the level of individual games. In this section we

briefly discuss the analyses of these decisions.

Ticket Quality. Most teams sell multiple price points of tickets. For example, if an area

has three sections, defined in terms of quality (A, B and C), it may be of interest to model

the selection of ticket quality tier. One approach to doing so could be an ordered logistic

regression model. This is similar to the binary logistic regression model used for

analyzing retention or initial buying decisions but the ordered logistic model allows for

multiple levels of a decision.

Given a set of customer characteristics, team characteristics and marketing decisions, it

becomes possible to predict the probability that a customer will purchase in section A, B

or C. However, the standard ordered probability model would likely need to be

generalized in order to model ticket quality. The critical issue is that multiple prices may

impact the section decision. For example in a 3 section arena where section A is the

premium section and section C is the least desirable the customer is confronted by three

distinct prices. The probability of selecting a seat in section A may be a function of the

prices in section A and section B only since customers interested in premium seats may

not even consider seats in section C.

It is also possible that it may not be possible for fans to migrate upwards to higher quality

tickets. If sections made up of higher quality tickets are usually sold out then a naïve

model would underestimate the desire of a fan to upgrade. In statistical terms, this is

known as censoring. For instance, during a cursory look at season ticket holder

decisions, we might find that customers who sit at the 40 yard line almost never migrate

to seats at the 50 yard line. An unsophisticated conclusion would be that these customers

are not interested in upgrading their seats. The problem is that higher quality seats may

not be available since seats on the 50 yard line are always sold out. In technical terms a

statistician would say the desire for upward quality transitions is censored or unobserved

due to capacity constraints. From a managerial standpoint, this type of situation may

obscure the true value of premium seats and may mistakenly lead analysts to conclude

that customer migration is a minor issue.




Page | 12

Modeling ticket quality decisions is an extremely important task. Given the level of

loyalty fans have for their preferred teams, season ticket holder relationships have the

potential to last for multiple decades. With customer lifetimes of this duration, it becomes

important to consider more general customer lifecycle issues such a general movements

towards higher quality seats.

Ticket Quantity. Another customer decision that may be analyzed using an ordered

probability modeled is ticket package size. Teams commonly offer complete season, half

season and other smaller season ticket packages. The natural size ordering of these

categories of packages means that an ordered probability model is again an appropriate

tool.

For example, if a team offered full, half, and quarter season packages, the ordered

probability model would be specified to predict the likelihood that a given customer

would select each level. As in the case of ticket quality, care would need to be taken in

terms of model specification. In particular, the manner by which price was included in

the model would need to be handled with care.

Purchase Timing. In our discussion of renewal decisions, we basically assume that

customer’s decisions are driven by variables that remain constant. This may be a

significant limitation if customer’s decisions are a function of variables that change over

time. For example, the probability of a renewal might change over the months of the off-

season as a team makes roster moves.

The engineering and medical fields have developed models for analyzing time until

failure or until death. These are called “hazard” models in engineering or “survival”

models in bio-statistics. The marketing field has begun to utilize these techniques to

predict when customer might make a purchase or end a relationship with a firm. As in

standard regression models, these techniques can include explanatory variables. An

example of when these models might be useful would be as a tool for analyzing the

impact of free agent signings on season ticket renewals.

Fan Expenditures. Teams may also have an interest in modeling total expenditures or

contribution. Sports fans tend to do a great deal of add-on spending beyond tickets.

Parking, concessions, and souvenirs may all represent significant contributors to the

value of a customer. Total expenditures is easily modeled using linear regression.

Standard regression models are appropriate since expenditures is a continuous variable.




Page | 13

It may also be useful to translate expenditures into margin or contribution. While the

marginal costs associated with a fan sitting in a seat are very low, the costs associated

with food or merchandise will be much higher. Therefore, it may be of great value to

convert revenues into margins since this would allow the CLV calculations to directly

speak to profitability. This is likely a simple task as multiplying category revenues

(concessions, tickets, etc…) by average category margins is probably sufficient.

However, while the analytical challenges may be simple, the integration of revenues from

different revenue centers may be a significant challenge.

Game Attendance Decisions. It may also be of interest to model game-level decisions,

such as whether a season ticket holder attends the game or sells the ticket on the

secondary market. At the level of the individual game, the analyst may use logistic

regression. The explanatory variables for this analysis might include opponent team

performance at that point in the season, recent performance trends, variables related to

the opponent, time of day, day of the week, weather and numerous other factors.

A game level analysis might be useful for a variety of business issues. For example,

understanding the impact of opponent characteristics, time of day, or whether the game is

played on a weekend could inform efforts to create a variable pricing schedule.

Understanding the relationship between customer characteristics and transaction history

variables might provide a basis for an early warning system for identifying at risk

customers.

Season level attendance could also be analyzed using a class of procedures known as

count models. A count model could be used to predict the number of missed games

based on individual customer factors such as ticket tier, distance to the stadium and years

as a season ticket holder. Customers that show significant deviations from the model

predictions would likely be very loyal or very much at risk.

Key Principles

While retention modeling is probably the core of customer analysis, there are

many additional decisions that should be analyzed.

Continuous outcomes such as total expenditure may be modeled using standard

linear regression, variables with an ordered structure such as ticket quality can be

analyzed using ordered probability models and purchase timing may be modeled




Page | 14

using hazard models. The point is that consumers make multiple types of

decisions and statistical techniques exist for each possible type of decision.

The value of modeling decisions beyond retention may be related to better

prediction of CLV or as input to decision tools related to identifying at risk

customers.

Many of the techniques identified above require specialized software and a high-

level of statistics training.




Page | 15

4. Ticket Quality and Pricing

A common challenge in modeling demand for sports is a difficulty in estimating the

relationship between price and demand. Game or section level models of demand as a

function of price will commonly yield positive price parameters. For example, if the best

seats are priced at $200 and these seats all sell out, while the most distant seats are priced

at $10 and often do not sell out, a naïve statistical model would suggest that demand

increases when prices are higher. The fundamental issue is that the seats within a

stadium or arena are of very different qualities. It is necessary to control for the

heterogeneous nature of inventory quality when modeling the relationship between price

and demand.

One approach advocated in the academic literature is to use the distance from some focal

point in the facility (home plate, mid-court, etc…). But this is often a flawed method.

The figure below shows a seating diagram for the University of Illinois’ Assembly Hall.

The circular shape of the seating area highlights the potential issue with using distance as

the quality metric. The distance to center court from for a given row in Section 7 will be

the same as a seat in Section 44 but it is doubtful that many fans would choose to sit

behind the basket if given a choice.




Page | 16

Thankfully, current trends are making available significant amounts of data that may be

used to create a meaningful ticket quality index. In terms of internal data, teams likely

possess an historical record that may be of value. Within a season, it is possible to look

at which seats sell first and which seats within a section sell last or remain unsold. The

growing secondary market contains a significant amount of information that may be used

to develop a ticket quality index. The most obvious data that can be extracted from the

secondary market is true reservation prices for each specific seat. In addition to

providing market prices for different seats, the secondary market can also provide

information on the timing of demand. Primary research methods such as conjoint

analysis could be employed to understand seat quality differences.

A seat quality index would be useful for estimating models of consumer demand. Rather

than use prices directly, it would be useful to create a dollar per quality index for sections

of similar seats.

The appropriate incorporation of price into CRM models varies based on the objective of

the analysis. In a simple binary renewal versus lapsed model, it may be sufficient to

include last season’s price and any price changes. However, the analyst should

understand that this approach includes an implicit assumption that customers do not

consider shifting up or down in terms of seat quality.

If the analyst wishes to include migration between quality (and quantity) levels, slightly

more sophisticated modeling techniques such as ordered logistic or probit models are

needed. Assuming that migrations tend to be to adjacent quality tiers then quality

adjusted prices and price changes for three classes of seats need to be included in the

model. This type of problem would require an ordered or multinomial probability model.

Key Principles

Incorporation of prices into season ticket and single game ticket sales is not a

straight-forward issue. The variation in quality of tickets across a stadium or arena

means that price elasticity cannot be properly assessed without an adjustment for

ticket quality.

Both internal and external data may be used to create an index of seat quality.

Internal data sources might include the order in which customers select seats

within a price point. The growing secondary market provides market level

willingness to spend on each seat.




Page | 17

Development of a seat quality index based on secondary market information

requires statistical models that control for factors such as time of day or opponent.

These analyses therefore can aid in the creation of variable and dynamic pricing

systems.




Page | 18

5. Customer Expectations

A critical aspect of consumer decision-making in the season ticket context is the role of

expectations. Considering consumer expectations introduces a great deal of complexity

into the analysis but this cost may be worth paying. The potential value in directly

modeling consumer expectations is that the link between on-field decisions and consumer

behavior is solidified and made explicit.

Expectations may take multiple forms. For this note we will consider two distinct types

of expectations. The more common of these is expectations of next season quality.

Expectations of next season’s quality are likely to be a function of last year’s

performance, any upwards or downwards trends from the previous season and the club’s

off-season additions or subtractions. For example, in response to the selection of Johnny

Manziel in the NFL draft, ESPN reported that the Cleveland Browns sold more than

2,300 season tickets within 24 hours of the pick. ESPN also reported that the Cavaliers

basically sold out of season tickets within 8 hours of LeBron James’ announcement that

he would return to Cleveland.

For a few teams, fans may have expectations regarding future scarcity. For example,

teams like the Packers or Steelers may endure losing seasons without loss of season ticket

holders because fans expect long waiting lists and limited access to re-subscribe. The

Packers report that the waiting list for season tickets “has more than 81,000 names” and

that the average wait for tickets is 30 years. If ticket quality is also a function of time as a

season ticket holder, fans may also have expectations regarding access to premium

tickets.

Modeling expectations is a tricky endeavor because expectations are not directly

observable. A variety of options are possible. One simple approach is to develop a

stand-alone performance forecasting model. Forecasted performance would then enter

directly as a covariate in the customer utility function.

A more complicated approach would be to estimate the model under the assumption that

customers act as dynamic optimizers. In this model, the customer’s sequence of

decisions is analyzed rather than just the immediate “one” season decision. These types

of dynamic programming models of customer behavior require a significant amount of

specialized knowledge and software.




Page | 19

Key Principles

Consumers, in general, and season ticket customers, in particular, often make

decisions based on expectations and other forward looking factors. To truly

understand season ticket holder behavior requires techniques that can incorporate

these expectations.

The types of expectations that are relevant will vary across clubs. Expectations of

next season performance are likely a factor for most teams. Expectations of future

season ticket scarcity are more applicable to teams with frequent capacity

constraints.

Techniques for modeling forward-looking behavior can range from including a

simple forecast of next season performance in a linear regression to developing an

explicitly dynamic statistical model that attempts to replicate the forward-looking

nature of consumer decision making.

Consumer expectations are obviously driven by off-season player moves. Explicit

modeling and inclusion of expectations in customer retention models therefore

helps to connect player decisions to marketing outcomes.




Page | 20

6. Optimal Customer Relationship Marketing

We now shift attention to the development of marketing and team strategy policies. The

figure below shows a simplified segmentation structure. The circles or nodes on the

figure represent four customer segments that are defined in terms of current and past

purchasing decisions.

The first segment is labeled “Prospect” and includes identifiable consumers that have not

made a purchase. In the case of the season ticket holders prospects could be defined in a

variety of ways. They might be limited to consumers that have made inquiries but not

made purchases, or the definition might be extended to include customers that have only

purchased single game tickets. The second segment is labeled “New Customer” and

includes first time season ticket purchasers. The third segment is labeled “Repeat

Customer” and is comprised of customers that have purchased in at least two subsequent

periods. The fourth segment contains customers that have previously purchased but have

failed to renew. This segment is labeled “Lapsed Customer.”

The arcs connecting the segments represent transitions between segments that occur

based on the annual decision of whether or not to purchase season tickets. For example, a

prospect may either make a purchase and transition to the first time buyer segment or not

make a purchase and remain in the prospect segment. New customers may renew and

become repeat customers, or fail to renew and become lapsed customers. Repeat

customers either renew and remain repeaters or become lapsed customers.




Page | 21

While this is a very simplified segmentation scheme, it highlights the team’s CRM

problem. Obviously, the key to maximizing value is to move customers from the non-

buying customer segments into the buying customer segments.

The arcs (transitions) should be viewed in terms of probabilities. In the specific and

limited segmentation structure illustrated in the diagram, the probabilities would be

derived from the customer acquisition and retention models discussed previously. A

critical point in this framework is that these probabilities are based on individual

customer traits, team performance and marketing decisions. The implication is that

marketing becomes at least partially responsible for the value of a team’s customer

relationship assets.

The final step in the CRM process is therefore how to develop marketing policies (and

possibly team spending policies) in order to maximize the value of current and future

season ticket holders. This might be accomplished through Monte Carlo simulations of

different marketing or team scenarios or through the application of dynamic optimization

techniques. For both of these techniques, the types of models of consumer response and

revenues we have discussed would be the primary inputs.

Key Principles

A critical element of CRM is to define customer segments in terms of customer

profitability. The goal of relationship marketing then becomes to drive customers

into higher profitability segments.

The retention and acquisition models discussed earlier should be designed to

predict the transitions between these segments or customer states.

Optimization procedures may be applied to determine the marketing and team

policies that maximize the team’s customer equity.

Optimization of customer management is a complex analytical problem.

Significant data and statistical expertise is needed to model customer behavior.

Dynamic optimization requires advanced skills in the area of applied mathematics

or operations research.




Page | 22

7. Final Comments

Our overarching goal in this note was to provide a structure for how teams (or more

generally any firm that operates in a capacity constrained environment) might approach

CRM. Along the way we have provided some guidance in terms of how various types of

consumer’s decisions might be modeled, and how these models can be combined to

develop improved marketing policies. The elevator version of the document would

involve the following three points:

Season ticket holders are valuable economic assets. Teams should view the

customers as assets and manage these customers in a manner that maximizes the

long-term value of each customer.

Trends in computing and information technology have progressed to the point

where it is increasingly possible for teams to collect the needed data to build

sophisticated models of a variety of consumer behaviors.

Data, statistical models, and optimization techniques may be combined to create

optimal marketing policies. In other industries it has been found that using

dynamic optimization to create individual level relationship marketing strategies

(such as targeted discounts or direct interventions) can result in 20% to 30%

growth in profitability.

One thing that we have tried NOT to do in this document is to overwhelm the audience

with technical detail. This was done for two reasons. First, we wish to make this article

(and future articles) accessible to a wide audience. As such, we have tried to keep a

balance between relationship marketing philosophy and statistical directions. Second,

managers in sports settings (and managers in just about any category) should be leery of a

one-size-fits all CRM and analytics solution. In our experience, there is a great deal of

need to customized modeling approaches and data collection across firms. For this

reason, we think it is more useful to discuss general approaches to modeling and

matching of techniques to problems rather than to go into detail.

a guide to season ticket holder management · a guide to season ticket holder management version...

Documents